The scientific merit of decadal climate projections can only be establ
ished by means of comparisons with observations. Testing of models tha
t are used to predict climate change is of such importance that no sin
gle approach will provide the necessary basis to analyze systematic er
rors and to withstand critical analysis. Appropriate observing systems
must be relevant, global, precise, and calibratable against absolute
standards. This paper describes two systems that satisfy these criteri
a: spectrometers that can measure thermal brightness temperatures with
an absolute accuracy of 0.1 K and a spectral resolution of 1 cm(-1),
and radio occultation measurements of refractivity using satellites of
the GPS positioning system, which give data of similar accuracy. Comp
arison between observations and model predictions requires an array of
carefully posed tests. There are at least two ways in which either of
these data systems can be used to provide strict, objective tests of
climate models. The first looks for the emergence from the natural var
iability of a predicted climate ''fingerprint'' in data taken on diffe
rent occasions. The second involves the use of high-order statistics t
o test those interactions that drive the climate system toward a stead
y state. A correct representation of these interactions is essential f
or a credible climate model. A set of climate model tests is presented
based upon these observational and theoretical ideas. It is an approa
ch that emphasizes accuracy, exposes systematic errors, and is focused
and of low cost. It offers a realistic hope for resolving some of the
contentious arguments about global change.